Early adopters of Grid computing have indicated that many are looking for ways to move beyond the use of grids specifically for computational tasks. They want to eliminate silos and organize IT around shared resources, simplify access to data, and provide a single, consistent view of the business for the entire enterprise. The challenge they face is essentially getting “beyond the compute grid,” a topic we've looked at in a number of reports.
Grid computing supports the key functional drivers required to achieve these goals, including virtualization, service-oriented architecture (SOA) and utility delivery models. To do this, grids need to be absorbed into the enterprise infrastructure itself and not used only as a stand-alone computational engine. In such a strategic role, grids must support automated data, storage and service activities just as capably as handling computational tasks. These challenges are being addressed by what is being called “Grid 2.0.”
If “Grid 1.0” is principally concerned with the virtualization, aggregation and sharing of compute resources, Grid 2.0 is focused on the virtualization, aggregation and sharing of all compute, storage, network and data resources. It is both service-oriented — uses Web services and provides access to IT as a service — and automated.
The 451 Group has tracked more than 250 early adopters of Grid computing and related technologies in the 451 Grid Adoption Research Service (GARS) and has developed a model to track deployment progress:
- Level 5 — Enterprise-wide grids, SOA, shared internal utility, outsourced utility.
- Level 4 — Linked grids, within or between departments and with multiple applications.
- Level 3 — Siloed grids: single grid, or grids in multiple departments that aren't linked.
- Level 2 — Single application running.
- Level 1 — Proofs of concept, trials.
This is not necessarily a sequential deployment path. Plenty of the organizations that 451 analysts have tracked are at level 2 or 3 and are not expecting to evolve through levels 4 and 5. Once organizations begin to link grids, their use can be thought of as relatively mature. At this stage, grids are well understood and often no longer the responsibility of a core R&D team. Once early adopters reach level 4, there are a range of level 5 strategic IT objectives that can be supported on top or outsourced.
Grid 1.0 usage is mostly at levels 1 through 3, and Grid 2.0 activities characterize levels 4 and 5. Financial services, pharmaceutical, electronic design automation, oil and gas, manufacturing and telecom vertical markets were the earliest adopters of grids. The applications they were and are using are primarily high-performance computing (HPC), embarrassingly parallel and batch-oriented in nature. Deployments are typically limited to one or two applications using traditional Grid middleware/schedulers. Culturally, they are very line-of-business oriented and in reality are mostly at level 2.
Grid vendors are seeking ways to expand beyond initial application beachheads. They are seeking additional early adopters within vertical markets beyond the early adopters, as well as application uses “downstream” of HPC.
In the Grid 2.0 era, use is extending across all verticals including retail, customer record management, insurance, travel, transport and government. Business intelligence, text analysis and search are some of the additional application tasks being supported on grids. Existing applications and application servers are also being deployed onto Grid-enabled containers without requiring prior knowledge of the infrastructure, and legacy applications are being wrapped for grids. Additional vendors are supplying tools to solve problems that are “beyond the compute grid,” while existing Grid vendors are diversifying to other markets and activities. Culturally, top-down, centralized approaches are driven by the CxO office and compliance is a key concern. The reality is that grids are being used to support new models such as shared internal utilities services.
So what are the kinds of activities for which grids are being deployed, and how do users describe their own deployments? When asked, 70 percent of early adopters who responded to a survey said there is a better term than “Grid” to describe their distributed computing architectures: 23 percent said virtualization, 23 percent said HPC, 19 percent said utility computing, 19 percent said clustering, and 15 percent said SOA.
Nothing Virtual about Virtualization
Virtualization enables emulation (makes something look like something else), aggregation (many things look like one thing) and segregation (one thing looks like many).
Is there a catch? Like grids, there is a good deal of confusion about what the term virtualization means, and two uses are emerging — virtualization inside a box and virtualization across multiple systems.
Furthermore, the effect virtualization will have on software licensing (and compounded by multi-core) has yet to be resolved. But it certainly will not be virtual. The major software vendors are terrified users will be able to extract more value from their software without paying any additional dollars. They're hoping that a mix of per-processor and per-user pricing mechanisms can satisfy users.
Who is Using Virtualization and Grids?
A leading European investment bank (that did not want to be named in this research) is using virtualization to create a “virtual resource market” that will allocate resources based on business demand, as part of its IT economy project. Other companies working on shared internal utility models using grids and virtualization include HSBC Bank, JPMorgan Chase, Royal Bank of Scotland, Bank of Montreal and BNP Paribas. Outside of financial services, users include Micro Technology Unlimited, Corus, DaimlerChrysler, GlaxoSmithKline and Arcelor.
Their challenges include the restrictions imposed by proprietary virtualization code, the ability to link grids over LANs and WANs and manage data and storage access, and how to create a virtual resource pool from a diverse range of IT assets and encourage the use of non-dedicated assets. HSBC's approach to the latter problem has been to offer services supported on shared assets at 50 percent of the cost IT will charge to groups requiring dedicated assets. One looming issue is the tax implications of hosting a grid in a different country from the users, which is not good from a users' point of view and indeed has forced HSBC to “park” its global shared grid strategy for the time being.
In the managed services and utility computing space (shared external utilities), the likes of Savvis Communications, Electronic Data Systems, Computer Sciences Corp., Hewlett-Packard, IBM, Oracle, Sun Microsystems, AT&T and BT are or expect to offer access to virtualized, shared resources. 451 analysts believe that there is a lot of marketing around utility computing, but not much of a market yet. In reality, there is little ability to pay only for what you use. Indeed, shared, internal utilities are today's hotspot.
Much like grids, SOA means different things to different people. What we do know is that you cannot buy an SOA. Different “ingredients” must be put together in order to build it. SOA is regarded as a way to implement services that are componentized, reusable and interoperable. The notion is that businesses using an SOA can be more competitive by responding more quickly to the changing market conditions through reuse and commodity services. It also potentially frees up internal IT resources to focus on innovating for the business rather than managing existing resources.
For the early adopters The 451 Group has interviewed, SOA is principally about the implementation of Web services. For them, Web services are afforded fault tolerance and scalability by running on a grid. Together with virtualization, this can deliver an SOA. Grids are a means to this end for many early adopters — especially in financial services. Grids are being regarded as the technology infrastructure and SOA as an application infrastructure.
451 analysts have noticed that SOA is principally concerned with new application development, while traditional HPC or compute grids are concerned mostly with chopping up existing applications. The bottom line is that grids allow users to run Web services better, faster and cheaper because they are cross-platform, can scale and offer better utilization. They also allow users to dodge platform and language wars.
Who is Using SOA and Grids?
At Bank of America, grids are seen as a corporate success story, although the grid itself is not an answer to anything. What is key is its ability to support service-oriented environments/SOA. The issue for Bank of America is that downstream data consumers cannot keep up with the volumes the grid generates. A plethora of support services are also required to keep grids primed, and resource ownership is an impediment to resource sharing.
Royal Bank of Scotland has a company-wide Web services (SOA) strategy, using grids for some part of it. Its big issue is that Web services equals federation which equals unreliability, it says. It is using grids and virtualization as the way around it.
Union Bank of Switzerland hopes to create an SOA that will support multiple teams and departments. It believes grids can support this. Its issue is that the value mostly resides in existing custom applications — and why should this be moved? It says Web services communication mechanisms are immature and that grids are not yet the tip of the spear for SOA.
A European investment bank (that did not want to be named) is creating a “virtual resource market” that will allocate resources based on business demand. It believes using SOA will allow IT to return to innovating for the business. Its issues are that data distribution can quickly become a headache, licensing for grids is a hairball and there will be vendor lock-in until standards shake out.
Other companies using grids to support an SOA strategy include Barclays Capital, Citigroup, Commerzbank, Deutsche Bank, ING Direct and JPMorgan. There are just as many that are specifically not going down the SOA route, including HSBC, TDS, BMO, DE Shaw, Markit Group and Bowne.
Initial conversations with early adopters revealed the business drivers for adopting grids were mostly concerned with price/performance (better utilization and optimization of resources, maximizing investments), cost sharing (multiple groups share cost of non-dedicated assets), and improved management (multiple systems under single management environment). If early adopters did not specifically mention saving money, they spoke of cost control and emphasized “hard” benefits.
Once grids have landed and expanded beyond beachheads, continued adoption has typically been framed by a very different set of business drivers. Collaboration and the linking of resources, being able to better respond to changing conditions, enabling shorter time to market and doing more new things are most often discussed. It suggests “making money” — and not only “saving money” — is the key here. HSBC reinforces this idea, saying “using grids doesn't save us money, but it will enable us to make more money.”
For more information, visit www.the451group.com/intake/gridtoday-gridworld-sept06.
About William Fellows
William Fellows is a principal analyst at New York-based The 451 Group — an independent technology industry analyst company focused on the business of enterprise IT innovation.
** GRIDwire, GRIDtoday's exclusive coverage from GridWorld 2006, can be seen at www.gridtoday.com/gridworld/06/index.html.